Executes and records auditable main research experiments against an accepted baseline.
npx skills add https://github.com/OpenLAIR/dr-claw --skill ds-experiment
Please help me install this Agent Skill. Skill: ds-experiment Source: https://github.com/OpenLAIR/dr-claw/tree/main/skills/ds-experiment Command: npx skills add https://github.com/OpenLAIR/dr-claw --skill ds-experiment
Running a research experiment is easy, but producing trustworthy, reproducible evidence is hard. This Skill enforces a disciplined run contract so every main experiment is comparable to its baseline, fully logged, and honestly reported.
Ask the agent to run the main experiment for the selected idea against the accepted baseline and record the results with metrics and a verdict.
High-intent search queries and answers about installing and using this skill.
Define a run contract first: hypothesis, baseline reference, dataset split, primary metric, stop condition, and budget. Then run a bounded smoke test, execute the full run with durable logging, and record metrics with an explicit supported, refuted, or inconclusive verdict.
A main experiment plan should state the research question, null and alternative hypotheses, baseline comparability rules, code touchpoints, smoke and full-run commands, fallback options, monitoring cadence, and a revision log for any route changes.
Do not start when the baseline gate is unresolved, the idea stage has open tradeoffs, or the evaluation contract is unclear. Resolve those through a decision step first, otherwise the run produces non-comparable evidence.
Launch the run as a detached managed session, monitor through durable logs with a wait-and-check cadence, and use structured single-line progress markers with ETA. Export the session log into the experiment artifact directory when recording results.
Comparability breaks when dataset splits, metric definitions, or evaluation code differ between runs. Verify the baseline metric contract, keep the baseline read-only, and record any evaluation change explicitly as supplementary rather than replacing the canonical comparator.
A complete run leaves a run manifest, metrics.json and metrics.md, summary.md, exported shell logs, config and command pointers, and a claim-validation record mapping each claim to its observed verdict and next action.
ResearAIStandardize a reference object for quest evaluation with plan and checklist requirements.
big4council-progValidate software ideas by building throwaway prototypes and issuing evidence-based verdicts.
ResearAIPolish research figures using a render-inspect-revise workflow and Matplotlib style.
verrio1Automate browser tasks with Playwright and capture console, network, and page errors.
